Engrammic MCP

Persistent, structured memory for AI agents.

Most agent memory is a bag of text chunks. Engrammic organizes memory into cognitive layers: observations become claims, claims become facts, facts become beliefs. Your agent doesn't just recall what happened; it knows what it learned and why it believes what it believes.

Quickstart

Install with one command:

curl -fsSL https://get.engrammic.ai/install.sh | sh

Windows (PowerShell):

irm https://get.engrammic.ai/install.ps1 | iex

The installer detects your agent harnesses (Claude Code, Cursor, Codex, GitHub Copilot, Gemini CLI, and more), wires Engrammic in as an MCP server, and optionally installs the Engrammic skills. Run engrammic-mcp login to authenticate after installation.

See Getting Started after installation.

To configure a harness by hand, add the MCP server directly:

{
  "mcpServers": {
    "engrammic": {
      "type": "http",
      "url": "https://beta.engrammic.ai/mcp/"
    }
  }
}

What It Does

Store something:

"Remember that the user prefers dark mode"

Recall it later:

"What do I know about user preferences?"

Trace why you believe something:

"Why do I think the user prefers dark mode?"

Tools

Tool	Purpose
remember	Store observations (no evidence needed)
learn	Store claims with evidence
believe	Form commitments grounded in facts
recall	Search and retrieve context
link	Connect related concepts
trace	Query provenance chains
reason	Record explicit reasoning steps
reflect	Store meta-observations
hypothesize	Form tentative beliefs
revise	Update tentative hypotheses
commit	Crystallize hypotheses to commitments
forget	Request node deletion
dismiss	Dismiss contradiction markers
patterns	Discover workflow templates
tick	Lightweight engagement check

Hosted vs Local

The default setup connects to beta.engrammic.ai. This is the right choice for most users.

Use the local engine (engrammic-engine) if you:

• Need fully offline or air-gapped operation
• Cannot send data to external services
• Are running a self-hosted production deployment

The local engine is a drop-in replacement. Point your harness config at http://localhost:PORT/mcp/ instead of the hosted URL.

Self-Hosting

For fully local, offline usage, see engrammic-engine.

Learn More

• EAG Paradigm - the cognitive architecture
• Skills - workflow guidance for Claude Code, Codex, Cursor, Gemini CLI

License

Apache 2.0